Census privacy changes will reduce detail for small places and population groups

For three months, the open question about the census privacy overhaul has been simple: what exactly gets lost, and when? The Census Bureau answered both this week, on its own website, under the byline of the official currently running the agency.

The answer on timing is that this stretches into 2029. The answer on cost is a single sentence the Bureau published in plain language: "Coarsening and suppression can reduce the amount and detail of data published, especially for small geographic areas or population groups."

That's the Bureau basically saying what outside researchers warned about in June and what its own officials conceded at a statistics conference in August. Small places will lose detail. We covered the original Commerce order and then the officials' confirmation, which came secondhand through notes taken by attendees. This time it's official text.

What changes, and when

The order, DAO 216-26, allows two ways to protect confidentiality. Coarsening, which means publishing less detail by grouping ages into ranges or rounding figures, is now the preferred method for everything. Suppression, which means withholding data outright, is allowed only as a last resort. Noise infusion, the family of methods that adds small random changes to published numbers, is banned, and that ban covers record swapping and synthetic data too.

Here's the Bureau's own product-by-product accounting.

Decennial data are "the hardest to protect due to the data's scale and detail," which is exactly why the 2020 count used the strictest version of noise infusion in the first place. The mid-2027 demonstration product is the moment to watch, since that's when data users get to see what a coarsened census actually looks like before the methods are locked in.

The accuracy argument, and the tradeoff

Adding noise to small-area statistics did make them wobbly, and demographers, redistricting analysts, and local planners spent years documenting cases where 2020 census numbers came back implausible for small towns and blocks. "While this approach was meant to protect confidentiality, it led to concerns that adding noise made the statistics less accurate and less trustworthy," the post says. The order's stated top priority is accurate statistics.

The catch is that the replacement methods don't make the confidentiality problem disappear. They solve it by publishing less. Coarsening trades detail for safety, and suppression removes numbers altogether, so the places carrying the highest re-identification risk, which are small towns, rural counties, census tracts, and small ethnic and language groups, are the same places where detail gets thinned. The Bureau acknowledges this and says it's "exploring ways to meet the data needs of these communities while protecting their information." What that looks like in practice isn't settled yet.

George M. Cook, listed as "Performing the Duties of the Director." The agency is making a generational change to how it protects data while operating without a permanent director, a gap we wrote about in the staffing crisis across federal statistical agencies.

What isn't affected

The order applies only to new products going forward, so everything already published stays exactly as it is. The 2020 Census and the 2024 American Community Survey aren't being rewritten or pulled. Internal microdata files that already have noise in them can still be used, as long as no new noise gets added to products built from them.

The order also doesn't reach independent research. A researcher working in a Federal Statistical Research Data Center, or a Bureau employee writing a journal article, conference paper, or working paper, can still use other disclosure methods as long as the work carries a disclaimer that the views are the author's own. So the academic literature keeps its tools even as the public data products lose theirs.

The Bureau says it's still writing the detailed implementation guidelines. Between now and the mid-2027 demonstration product, the practical question for anyone who relies on small-area statistics is which tables survive coarsening, which get suppressed, and which quietly stop appearing.

Sources

The policy details, timeline, and all quotes are from the U.S. Census Bureau director's blog post, Understanding the New Disclosure Avoidance Policy, published August 17, 2026 by George M. Cook, performing the duties of the director. The underlying order is Department Administrative Order 216-26, issued by the Department of Commerce in June 2026. Implementation guidance is being developed in the Bureau's DS-025 data stewardship policy. Our earlier coverage of the order and the Bureau's August confirmation is linked above.